Download The F14A Central Air Data Computer and the LSI Technology State

The F14A Central Air Data Computer
and the
LSI Technology State-of-the-Art in 1968
Ray M. Holt
updated September 22, 1998
This paper is designed to be a supplement to the paper “Architecture Of A Microprocessor”
written in 1971 by Ray M. Holt. This paper is written because of the recent declassification
of the 1968 F14A Central Air Data Computer (CADC) microprocessor design.
Introduction
With 30 years of computer architecture design, MOS/CMOS process technology, and
computer history written in stone, it is difficult, at best, for the current readers to critically
examine the architecture and technology decisions made in the time frame of 1968-69. I will
attempt to set the stage with what we faced in 1968 for technology choices and I will attempt
to present some of the design trade-off’s that were made that time. The architecture and logic
of the F14A MOS-LSI chips were designed my Mr. Steve Geller and myself.
Of course, it is normal and natural for the current computer designer to compare all designs
with the current Intel series of microprocessors. Intel has accomplished a tremendous feat in
producing commercial microprocessors that have been able to meet or exceed the expectation
of the computing world. However, the current accepted microprocessor definition, (i.e. the
single-chip CPU) is a by-product of, mostly, Intel’s marketing needs and commercial motives
and, not necessarily, the best integration of computing architecture and technology. The
commercial and consumer world uses what is available at a cost-effective price.
What Did The CADC do?
The CADC performed the function of controlling the moving surfaces of the aircraft and the
displaying of pilot information. The CADC received input from five main sources, 1) static
pressure sensor, dynamic pressure sensor, temperature sensor, analog pilot information, and
digital switch pilot information. The output of the CADC controlled the moving surfaces of
the aircraft. These were the wings, maneuver flaps, and the glove vane controls. The CADC
also controlled four cockpit displays for, Mach Speed, Altitude, Air Speed, and Vertical
Speed. The CADC was a redundant system with real-time self-testing built-in. Any single
failure from one system would switch over to the other. Two state-of-the-art quartz sensors, a
20-bit high precision analog-to-digital converter, a 20-bit high precision digital-to-analog
converter, the MOS-LSI chip set, and a very efficient power unit made up the complete
CADC. A team of over 25 managers, engineers, programmers, and technicians from
AiResearch and American Microsystems labored for three years to accomplish a design feat
never before attempted, a complete state-of-the-art, highly integrated, digital air data
computer. This paper will discuss the MOS-LSI microprocessor chip set only.
© 1998 copyright Ray M. Holt ALL RIGHTS RESERVED
1
The Problem
Previous Central Air Data Computers were mechanical designs. Gears, cams,
potentiometers, etc were designed to implement the algorithms for the mathematical
functions necessary to control the aircraft. With the F14A, a newer, lighter, faster, and higher
performance aircraft was needed. Mechanical computer were not only too heavy, but they
required extensive design changes to implement any changes in operational function.
The challenge presented to the F14A CADC design team was to develop a small, lightweight,
reliable, low power, rugged, high performance, and flexible computer that could be shipped
within two years. Extensive monetary penalties were placed on not meeting the design and
delivery goals.
The F14A CADC computer had the following design requirements:
• Must fit onto one 4” x 10” printed circuit card, 40 square inches
• It’s power must not exceed 10 watts at room temperature
• Must be flexible enough to handle computing capacity changes during the
development cycle
• Must be programmable to accomplish eventual functional changes
• Must be able to self-test and report any failure
• Must operate at high-G forces and at Mach 1 speeds
Design Considerations
The choices for technology in 1968 was either TTL bipolar, MOS chip modules (and, or
gates and flip-flops), or large-scale integrated (LSI) circuitry.
TTL bi-polar technology was eliminated because of the high package count required and for
the high power consumption required. MOS gate modules were eliminated because of the
high package count and the space required to implement the computing requirement.
If we were to meet the necessary requirements we were forced to consider, and eventually
use, the new state-of-the-art P-Channel MOS technology. This technology being only a year
or so old was attractive because of its high density capability, its low power requirements,
and hopefully its low, or reasonable, cost. Eventually, American Microsystems Inc of Santa
Clara, CA was our selected as our preferred vendor because they had custom chip design
experience, was willing to work with us, was able to provide extremely capable design
engineers, and had proven capability in the P-MOS process technology. Even so, this project
proved to be very aggressive considering its time schedule, its mil spec temperature
requirements which greatly reduced the chip densities, and the computing architecture styles
which had never been integrated at such a large scale.
© 1998 copyright Ray M. Holt ALL RIGHTS RESERVED
2
Large Scale Integration – 1967-69
I will present the state-of-the-art at the time of the F14A CADC microprocessor development
by referring to several papers, articles, and documents I used during the design trade-off’s
and one’s I have obtained in recent months.
Joseph, Earl C. Impact of Large Scale Integration on Aerospace Computers IEEE
Transactions on Electronic Computers, Vol EC-16, No. 5, October 1967
Abstract
“The continued development of Large Scale Integration (LSI) presages the advent of a fourth
generation of computers, and is causing an upheaval at all levels of computer technology – both
technical and managerial. Today’s computers use integrated circuit components containing at
most ten gates per component; however, LSI is introducing hundreds of gates per component
and will eventually evolve into thousands of gates per component. To the aerospace planner,
this technological breakthrough of LSI means tremendous reductions in cost, size, weight, and
power consumption of logic components, together with increased speed and reliability.
However, for the aerospace planner to successfully implement aerospace computers with LSI,
the computer designer and manager must reorient their methodology and goals. A multitude of
new design and cost considerations must be carefully scrutinized, and out of this must come the
new techniques that will permit effective incorporation of LSI in aerospace computers.
Higher speed, a smaller system, greater reliability, and lower cost stem from the physical
structure (more gates per component with no increase in component size) and the LSI
component. These inherent features of LSI, together with multiprocessor system organization,
point to future aerospace computers with capabilities equal to today’s best ground base
systems.”
“Today, we build our computers using integrated circuits components with two or four and at
most ten gates per component.”
Pin Minimization
“The designer will be forced to organize the logic per component in a fashion that will minimize
the number of pins required to connect the LSI into the system.”
Reliability
“It is of great importance to the aerospace system designer that the LSI features of higher speed
and smaller systems can be obtained with greater reliability, lower power, and at a reduced cost
over present day systems.”
Parity (error) checking
“In an LSI component, where many logic nodes exist, multiple logic node failures are likely,
such as, when a crack propagates through a chip.”
“Most likely, in the future the designer will resort to forms of multiple checking (various coding
techniques) with LSI.”
© 1998 copyright Ray M. Holt ALL RIGHTS RESERVED
3
Survival of Industry in the LSI Era
“LSI protends the computer on a chip – would you believe a single wafer or perhaps two
wafers? This certainly will occur; the only question is when.”
Conclusion
“We are only at the threshold of LSI. Its large-scale implementation lies ahead of us.”
Levy, Saul Y. System Utilization of Large-Scale Integration IEEE Transactions on
Electronic Computers, Vol EC-16, No. 5, October 1967
“The system architectural problem is one of maximizing the benefits of reliability and speed to
be derived from utilizing the LSI technology, benefits obtained by minimizing connections
external to the arrays and encompassing all functionally related logic in one tightly packed
array.”
Feth, G.C. & Smith, M.G. “Large-Scale Integration Perspectives” Computer Group News,
November 1968
Introduction
“LSI is a physical reality, yet it has a long way to go to reach the impact so often forecast.”
“How large is large-scale is not fully agreed upon. Is it 50, 100, or 200 circuits on a chip; is it
a computer on a wafer; or is it simply a level of integration just a little beyond what is
attainable today?”
LSI is Packaging
“An important point which should be emphasized here is that the economic impact of LSI is not
only on the basic circuit costs, but also in the displacement of packaging costs. In a real sense
LSI is packaging.”
MOS Technology
“Consequently, the simple geometry of MOS circuitry leads to very high packing density, which
is particularly appropriate to LSI.”
Larger and Cheaper
“Table II presents a representative projection of MOS integrated circuitry.”
© 1998 copyright Ray M. Holt ALL RIGHTS RESERVED
4
Packaging
“Of all the claim made of LSI, few have promised the complete system-packaging efficiency
representative of an advanced technology. While concentrated efforts are being made to produce more
complex chips (with high device densities and higher yields), the industry patiently awaits the time when
it can specify and purchase a packaged computer on a slice. This goal may indeed be desirable, but
major problems still intervene. In the meantime, most companies are attempting to optimize parts as
well as the user-designer interface at the chip level.”
LSI for Memory
“A major LSI interest today is in memory. Obviously, it would be a significant boon to LSI if it could
compete for the memory market, and particularly the main memory market.”
Conclusion
“LSI is a natural and fundamental extension of technology. There are variations, but ultimately the
direction is inescapable.
“LSI imposes some additional conditions in its implementation and application. We will have to be a
little more clever.”
Fischel, Jack & Gee, Shu W. Aeronautical Flight-Control Systems Research Aerospace
Electronic Systems Technology: A Briefing for Industry, May 3-4, 1967, p 259
“The primary flight-control system configuration now being used, an electrical-mechanical-hydraulic
combination, has proved to be fairly reliable… However, the rate of advancement of the fly-by-wire
system rest entirely with the electronics industry in providing completely reliable, low-cost, easily
maintained systems for the more exacting flight control of future aircraft.”
Boyce, Jefferson C. Microprocessor and Microcomputer Basics Prentice-Hall 1979 p. 3
“Continued development and new techniques finally resulted in capabilities to pack hundreds of
components into extremely small spaces, and in about 1967 the first practical LSI assembly emerged. It
was this high-density packaging that has allowed the microprocessor to become a reality and to usher in
the era of low-cost computing.”
Computing Architecture
Alt, Franz L. & Rubinoff, Morris Advances in Computers Academic Press, Vol 9, 1968, p.17
New Systems
“Most computing systems of the last two decades have had a fundamental structure proposed by von
Neumann, i.e. they consist of a certain number of registers connected to a memory on one hand and to an
arithmetic unit on the other with appropriate shuffling of numbers dictated by a control unit. It is a
tribute to von Neumann’s genius that this arrangement should have become completely standard. The
time has come, however, for compute designers to reassess the situation and to think in terms of
arrangements in which either the memory and processing functions are more intimately connected or in
which there is a very strong parallelism even to the extent of having hundreds of arithmetic units.”
© 1998 copyright Ray M. Holt ALL RIGHTS RESERVED
5
Selecting a computer design architecture is like selecting the best house for a given piece of
property. There are many parameters to consider when deciding upon the optimal configuration.
For computers these parameters are usually based upon speed, size, power consumption, and
cost.
When designing without size and power consumption constraints computing architecture
designs are centered around speed. How fast can we make this machine go? How can we
optimize our instruction set to pick up speed? When small space and low power are included
into the design constraints then the overall problems is complicated ten-fold, at best.
Some of the issues we were faced with in determining the final F14A CADC microprocessor
architecture are:
• PC board space – will it all fit on one PC card of 40 square inches?
• Computational speed – can we perform all functions in the required time, including the
single-failure built-in self-test?
• Reliability (based on pin count) & technology stability
• Project schedule & the likelihood of finishing on time
• Power consumption – can we keep it at 10 watts at room temperature
• Temperature range – will it operate over mil spec temperature ranges
• Will it withstand the acceleration and mechanical shock?
Given these constraints we ended up with the following design guidelines:
• P-MOS technology would be used because of its high density capability
• Unique chip designs would be kept to a minimum
• Package size would be kept to a minimum
• Package quantity would be kept to a minimum
• Package pin counts would be minimized overall
• Data transfer between any chips or computing modules would be serial
Working within these guidelines we had to come up with a computer architecture that would
meet the performance requirements.
Our first approach was the classical single arithmetic unit, instructions stored in ROM,
temporary values and results stored in random-access memory and some form of
input/output. This well know design which could have been implemented easily with the
current MOS technology did not have enough computational power. In other words, we
could not drive the chips fast enough, at the required temperature spec’s, to perform all of the
necessary functions. If we were implementing a consumer application, such as a calculator,
at consumer spec’s, it would have been accomplished with ease.
© 1998 copyright Ray M. Holt ALL RIGHTS RESERVED
6
Our second approach was to come up with a way to perform computations in parallel. In
order to accomplish this we had to analyze the functions of the aircraft to see if and how
much of the operations can be performed at the same time without causing any dynamic or
stability problems to the aircraft control surfaces. An exhaustive review of the mathematical
functions and the dynamics of the aircraft allowed us to come up with a parallel, multiprocessor implementation whereas we could have 2 or more CPU units running at the same
time. In order to accomplish this we had to synchronize the CPU’s and to provide an
instruction command set that allowed data transfers between each.
Additionally, it was decided to increase the throughput of each CPU by providing an
additional register, whereas, we could “pipeline” the instruction and data transfers. This
meant that while new data was being calculated, old data was being shifted to another CPU
or into random-access storage. This “pipelining” provided a tremendous speed increase
without adding much to the chip space. The final configuration was determined after many
weeks of design and consultation with the engineering team at American Microsystems. The
final architecture was a 20-bit, multi-purpose, microprogrammed, pipelined, multi-processor,
using MOS P-channel technology. Only six unique chips were required. The total number of
devices (or transistors) in the system was 74,442 of which 62,092 was the stored program
instructions in read-only memory.
Pushing the limits of the technology and the complexity of the architecture put an additional
risk factor on the project schedule. We had less than one year to implement the chips,
program the ROM’s, fabricate the chips, and test the system. In order to reduce this risk
factor as much as possible we used a team of programmers to entirely simulate the chip
design (transistor by transistor), the timing between the chips, and the stored program
instructions. Before we committed to final silicon with the chips we had a 99% confidence
level that everything would work. We were not so brave as to expect 100%, even though we
would have not been surprised had we accomplished it.
In addition, we had the requirement to completely self-test all the chips in the system and
report any single failure. In order to do this we used our program simulation and generated
an exhaustive set of test patterns and results that would continuously run through all of the
chips along with the normal flight requirements. Our simulation showed us we accomplished
this goal.
The completed F14A CADC was prototyped with off-the-shelf MOS devices and
demonstrated to operate as designed. After extensive software simulations, test patterns,
prototyping, we committed our instruction set to ROM. Turn around on ROM was several
months. We anxiously waited the arrival of the first ROM set and plugged them into our
actual first CADC system. The first day of testing was 100% successful. Every function
performed flawlessly. The second day we discovered a slight, but critical, glitch in moving
one of the surfaces. Our investigation resulted in one ROM bit was incorrectly programmed.
One out of 62,092 is a great percentage, especially for hand programming, however, it
required us to wait another month or so for the new ROM. The second set was perfect.
© 1998 copyright Ray M. Holt ALL RIGHTS RESERVED
7
Did It Work?
First CADC’s were delivered to Grumman in early 1970. The first flight of the No. 1
prototype F14A took place on December 21, 1970. The fully proven F14A was introduced to
the fleet only 51 months after contract award. A total of 557 F14A’s were produced of
which 32 are still scheduled for upgrades to F14A+. The Tomcat's size, speed, and range,
outstanding cockpit visibility, coupled with its unparalleled radar and missile systems makes
it the premier fighter in Western service according to many observers. Grumman succeeded
in delivering the F14A on time, on cost and as an even better fighter than they contracted for!
Moore’s Law
Gordon Moore’s observation, in 1965, of semiconductor growth and performance led to what
is now called Moore’s Law. Each new chip contained roughly twice as much capacity as it
predecessor, and each chip was released within 18-24 months of the previous chip.
Applying Moore’s Law to the F14A microprocessor chips which were produced
approximately 18 months prior to the Intel 4004 meant that we should have only been able to
put 1,000-1,500 devices (or transistors), or less, on a chip when, in fact, we were able to get
as much as 3,268 and at mil spec temperature ranges. A testimony to the engineers and the
P-channel process at American Microsystems Inc.
Micro Is The Name
In the ‘67-‘69 timeframe, as technology was advancing towards 100’s and even 1000’s of
devices (transistors) per chip several terms starting appearing describing this new breed of
technology.
Integrated Circuits
LSI – large-scale integration
MOS-LSI large-scale integration made from the MOS technology
Microelectronics – small electronics or more specifically logic gates in one package.
Microminature – small circuit boards containing complete functions normally on several
large circuit boards. Mostly accomplished using LSI circuits.
Computer-on-a-chip (or slice) – intended to mean the entire computer on a chip, however,
nobody really considered the I/O as part of the computer which even to this day would not
make a computer-on-a chip. Later, the arithmetic unit on a chip. Even later, the central
processing unit, or CPU, on a chip.
© 1998 copyright Ray M. Holt ALL RIGHTS RESERVED
8
Microcomputer – Initially, in 1968, a computer implemented on several circuit boards. Later,
on one circuit board. Even later, on several chips.
Microprocessor – Initially, in 1969, the processing unit of a computer on a single-card.
Later, an arithmetic unit on a chip. Even later, the ALU with instruction decoding and a few
registers on a chip. This later definition was marketed by Intel and became a “standard”
definition around the 1973-74 timeframe.
What is A Microprocessor?
Prior to the commercialization of the LSI computer chips this question was not of too much
concern to most people except chip designers.
However, this is 1998 and many are interested in whether the F14 CADC microprocessor
was, in fact, of similar definition to the 4004 built 1-2 years later.
•
1973 from Hank Smith, Microprocessor Marketing Manager, Intel Corp.
IEEE 1973 WESCON Professional Program Session 11 Proceedings
“A CPU uses P-channel MOS and is contained in 1, 2, 3 or 4 LSI standard dual-in-line
packages from 16 - 42 pins per package".
• Intel’s definition off its web site at:
http://www.intel.com/intel/museum/25anniv/Hof/whatis.htm
says, “A microprocessor is an integrated circuit built on a tiny piece of silicon. It contains
thousands, or even millions, of transistors, which are interconnected via superfine traces of
aluminum. The transistors work together to store and manipulate data so that the microprocessor
can perform a wide variety of useful functions. The particular functions a microprocessor performs
are dictated by software.”
• Boyce, Jefferson C. Microprocessor and Microcomputer Basics Prentice-Hall 1979
“A microprocessor is, by generally accepted definiiton, a large-scale integrated circuit assembly that
contains about 70% of the computing capability of a very small digital computer.”
Certainly the F14A microprocessor met these definitions. However, to apply a definition to a
design before the definition was even in existence is slightly short-sighted and certainly takes
away from the accomplishments of the F14A microprocessor architecture and chip design.
© 1998 copyright Ray M. Holt ALL RIGHTS RESERVED
9
My Opinion
The fine line between a small minicomputer, a microcomputer, or a microprocessor was quite
a mute point for many years as the quantity of units that would qualify under these
definitions was very small and the number of people who cared was even smaller. As the
commercial market for small computing devices began to explode the definition marketed by
Intel dominated.
Consequently, the perception today is that small computers, the MOS-LSI computers, and
other highly integrated devices did not exist prior to 1971 i.e. the introduction of the 4004.
When in fact, 1971 was almost two years after the initial design of the first MOS-LSI
microminature, microprogrammed, parallel processing, pipelined computer, which I will take
the liberty, as designer, to call it the very first microprocessor, because that is, in fact, what
we called it. A significant achievement, under very tight constraints, by a tremendous team,
for a world class aircraft.
My Claims for the F14A Microprocessor and Related Chip Set
1st microprocessor and microcomputer
1st aerospace microprocessor and microcomputer
1st “fly-by-wire” flight computer
1st military microprocessor and microcomputer
1st production microprocessor and microcomputer
1st fully integrated chip set microprocessor and microcomputer
1st 20-bit microprocessor and microcomputer
1st microprocessor and microcomputer with built-in programmed self-test and redundancy
1st microprocessor and microcomputer in a digital signal processor (DSP) application
1st microprocessor and microcomputer with execution pipelining
1st microprocessor and microcomputer with parallel processing
1st integrated math co-processors (PDU and PMU)
1st Read-Only Memory (ROM) with a built-in program counter
Respectfully submitted,
Ray Holt
© 1998 copyright Ray M. Holt ALL RIGHTS RESERVED
10